SCH: Integrating AI and System Engineering for Glucose Regulation in Diabetes

点击翻译按钮获取中文摘要

The objective of this proposal is to develop a prototype for the next generation multivariable automated insulin delivery (mvAID) systems (also called artificial pancreas) by integrating systems engineering and artificial intelligence (Al) techniques that will mitigate the effects of meals, physical activities ,acute psychological stress inducements and sleep irregularities without manual inputs by the user to tightly regulate the glucose levels of people with diabetes. The first generation of automated insulin delivery (AID) systems relied on hybrid closed-loop technology, collecting data from continuous glucose monitoring devices and requiring manual user inputs for mitigating the effects of meals and exercise. The multivariable AID that we developed provides a well-integrated next-generation system that analyzes historical and realtime data from different sources, including continuous glucose monitoring systems, insulin pumps, and wearable sensors in wristband physical activity trackers, to mitigate the effects of meals, physical activities, and acute psychological stress without manual inputs by the user. Meals, planned exercises, many physical activities of daily living, acute psychological stress, and sleep irregularities affect blood glucose levels differentially, challenging people with Type 1 diabetes to continuously consider all these complex factors in maintaining their blood glucose levels in the target range. Further improvement in glucose regulation can be achieved by developing novel, interpretable, and interactive Al techniques that can explain their predictions to medical care providers and AID users, and by integrating these Al techniques with systems engineering techniques to develop an Al-mvAID system. The function of these Al techniques is to predict the state of a person based on historical trends and current data, and provide additional valuable information to the mvAID system to relieve the users from onerous repetitive tasks for interpreting their current metabolic state, predicting the impact of their current actions on future variations in glucose levels, and tuning the parameters of the Al-mvAID controller. The goal is to produce a powerful userfriendly technology that integrates novel Al techniques with mvAID systems for minimal user burden in achieving tight control of glucose levels despite the many complex glycemic disturbances occurring in freeliving conditions, such as meals, physical activities, acute psychological stress, and sleep irregularities.